WMOD00563

• Title: Snow Accumulation Algorithm for the WSR-80D Radar: Final Report
• Date: 7/1/1998
• Weather Modification - Doc Type: Report

Another likely reason for optimization scheme failure with some individual gages is limited sampl1e size.
Krajewski and Smith (1991) show the need for large numbers of data pairs in their simulation of
synchronous observations. Ideally, hundreds of pairs of observations would be used to establish Ze-S
relationships. Table 4, where all gage or snow board observations within about 60 km range are grouped,
provided much larger sample sizes than table 3 for individual surface sites. Probably as a result, the p
values only range from 2.0 to 2.3 in the grouped site results of table 4 (with the exception of Minnesota
which had the fewest pairs and the smallest range of hourly S accumulations).

Table 4.-Summary of a and 13 values resulting from application of the optimization scheme to
combined data sets within about 60 km of each radar. The minimum hourly S accumulation usecl was
0.005 inch for all gages and 0.100 inch for snow boards. Accumulation of S was set to zero for 2~
values less than 5 dBZ.

Surface No. of a for
Radar 10 Site 10 beam tilt hours a 13 13 = :2.0
KCLE 1,2 0.5 698 260 2.0 260
KENX 20 boards 0.5 1009 108 2.3 117
KFTG 1,2,3,7 0.5 458 135 2.2 132
KGJX 1,2,3 2.4 1137 42 2.0 42
KMPX 1,2 0.5 354 240 2.6* 184
* KMPX did not optimize

Examples of how CTF, normalized by dividing by the sample size, varied with P are given in figures 4a
and 4b for the grouped data of table 4. This figure shows that relatively large CTF values exist in aU
cases for P less than about 1.6. This finding indicates it is unlikely that P is less than about 1.6 for
snowfall. The minimum CTF values for 4 of the 5 plots indicate that P is approximately 2.0 although the
minima are rather board even with these relatively large sample sizes.

The primary optimization scheme worked well for individual gages in regions with a wide range of
hourly accumulation like the top of the 3 km elevation Grand Mesa (KGJX) and the nearest 3 gages
across the lake effect snow belt east-northeast of Cleveland (KCLE). Of course, these regions of
frequent snowfall also produced 300-450 sample pairs at individual sites in contrast to the other regiions
where 100-200 h of snowfall was typical. Failure to optimize for the 2 most distant KCLE gages is
likely related to erratic Ze measurements at ranges beyond 100 km. These could be caused by the lowest
tilt radar beam sampling the upper portions of the shallow clouds and even sometimes overshooting cloud
tops.

6.4 Alternate Scheme

In cases where the primary optimization scheme failed to find a CTF minimum in curves such as
for KMPX in figure 4a, an alternative approach was used. This approach did not require either absolute
equality of the hourly mean values of surface-observed and radar-estimated S, or that a minimum CTF
value from equation (7) be achieved. Instead, the difference in hourly means was calculated for many
combinations of a and p. An iteration process was used in which a was stepped by 10 and p was stt~pped

25